The Multifunctional Pipette
A Microfluidic Technology for the BiosciencesALAR AINLA
Department of Chemical and Biological Engineering
CHALMERS UNIVERSITY OF TECHNOLOGY
ALAR AINLA
ISBN: 978-91-7385-818-2
©Alar Ainla, 2013
Doktorsavhandlingar vid Chalmers tekniska högskola
Ny serie nr: 3499
ISSN: 0346-718X
Department of Chemical and Biological Engineering
Chalmers University of Technology
SE-412 96 Göteborg
Sweden
Telephone: +46-(0)31 772 1000
Front cover image: Multifunctional pipette -- a toolbox for single-cell
and biomembrane studies. You may guess which tool represents which
experiment in the papers!
Back cover photo by Viktoria Gusak
Printed by Chalmers reproservice
Göteborg, Sweden 2013
ALAR AINLA
Department of Chemical and Biological Engineering Chalmers University of Technology
Abstract
The theme of the work described in this thesis is the generation and application of liquid microenvironments in chemistry and bioscience using microfluidic devices. First, a computer controlled multi-stage dilution system to generate time-dependent chemical waves was developed, and its application was demonstrated on model biomembranes. Thereafter the focus was shifted towards spatial control of chemistry. Using a hydrodynamic flow confinement concept in an open liquid volume, we created a device coined “Multifunctional Pipette”. It features localized liquid handling at the single-cell size scale together with fast solution exchange. The technology has been refined and optimized to provide a feature-rich tool for biologists working with cells and tissues in microscopy experiments. Application examples include cell zeiosis, single-cell dose-response determination and ion-channel stimulation. Subsequent studies cover modifications and applications of this device, such as on-chip electrodes and electroporation, as well as uses in cell cultures, on tissue slices, and as an optofluidic thermometer. Finally, localized liquid handling has been applied to assemble 2-dimensional fluidic networks consisting of directly written supported lipid bilayers. This "Lab on a Membrane" toolbox allows rapid prototyping of 2D-fluidic circuits, to modify their chemistry and connectivity on-demand and to apply them in studies of molecular interactions.
Keywords: Microfluidics, microfabrication, PDMS, microfluidic dilution, microfluidic mixer, hydrodynamic flow confinement, microfluidic superfusion, single-cell analysis, supported lipid membranes, microfluidic temperature sensing.
I
A Microfluidic Diluter Based on Pulse Width Flow Modulation Alar Ainla, Irep Gözen, Owe Orwar & Aldo JesorkaAnalytical Chemistry 2009, 81(13), 5549-5556.
II
A Microfluidic Pipette for Single-Cell PharmacologyAlar Ainla, Erik T. Jansson, Natalia Stepanyants, Owe Orwar & Aldo Jesorka
Analytical Chemistry, 2010, 82(11), 4529-4536.
III
A multifunctional pipetteAlar Ainla, Gavin D. M. Jeffries, Ralf Brune, Owe Orwar & Aldo Jesorka
Lab on a Chip, 2012, 12(7), 1255-1261.
IV
Single-Cell Electroporation Using a Multifunctional PipetteAlar Ainla, Shijun Xu, Nicolas Sanchez, Gavin D. M. Jeffries & Aldo Jesorka
Lab on a Chip, 2012, 12(22), 4605-4609.
V
A multifunctional pipette for localized drug administration to brain slices Aikeremu Ahemaiti, Alar Ainla, Gavin D. M. Jeffries, Holger Wigström, Owe Orwar, Aldo Jesorka & Kent JardemarkManuscript
VI
An optofluidic temperature probeIlona Węgrzyn, Alar Ainla, Gavin D. M. Jeffries & Aldo Jesorka Submitted manuscript
VII
Lab on a Membrane: a Toolbox for Reconfigurable 2D Fluidic NetworksAlar Ainla, Irep Gözen, Bodil Hakonen & Aldo Jesorka Submitted manuscript
I
Proposed concept for the diluter. Designed and fabricated microfluidic devices. Designed pneumatic and electronic control system and software. Performed calibration and testing. Analyzed data and developed models. Participated in lipid spreading experiments. Designed figures. Contributed to the writing of the paper.II
Proposed the pipette concept. Designed and fabricated all microfluidic devices. Designed control mechanism and software. Performed calibrations and testing. Performed finite element modeling. Analyzed data. Participated in all biological experiments. Designed figures. Contributed to the writing of the paper.III
Contributed to the design of the pipette shape and interfacing. Designed and characterized microfluidic circuitries. Participated in switching speed experiments. Performed calculations and developed models. Performed biological experiments. Wrote the paper.IV
Participated and supervised electrode development. Designed electrical interface. Participated in all biological experiments. Performed calculations and developed models. Designed figures. Contributed to the writing of the paper.V
Participated in intracellular recording experiments. Designed figures. Contributed to the writing of the paper.VI
Proposed concept of dye multiplexing. Developed models and performed simulations. Participated in experimental planning. Participated in all experiments. Contributed to data analysis. Designed figures. Contributed to the writing of the paper.VII
Proposed concepts of writing and erasing lipids using multifunctional pipette. Participated in experimental planning. Wrote control software. Participated in all experiments. Performed simulations. Analyzed and interpreted all data. Designed figures. Contributed to the writing of the paper.Thermal Migration of Molecular Lipid Films as Contactless Fabrication Strategy for Lipid Nanotube Network
Irep Gozen, Mehrnaz Shaali, Alar Ainla, Bahanur Ortmen, Inga Põldsalu, Kiryl Kustanovich, Gavin D. M. Jeffries, Paul Dommersnes, Zoran Konkoli & Aldo Jesorka Submitted manuscript
Calibrated On-chip Dilution Module for the Multifunctional Pipette Andreas Genner, Alar Ainla & Aldo Jesorka
Proceedings of the 3rd International Workshop on Soft Matter Physics & Complex Flows. Ed.: Jon Otto Fossum and Elisabeth Bouchaud. Lofoten, Norway
Influence of Temperature on Enzyme Activity in Single Cells
Shijun Xu, Alar Ainla, Gavin D. M. Jeffries, Kent Jardemark, Owe Orwar & Aldo Jesorka
Manuscript Book chapter:
Hydrodynamically Confined Flow Devices Alar Ainla, Gavin D. M. Jeffries & Aldo Jesorka
In book “Hydrodynamics – Theory and Model”, by Jinhai Zheng, InTech, 2012 ISBN 978-953-51-0130-7
Review article:
Hydrodynamic Flow Confinement Technology in Microfluidic Perfusion Devices Alar Ainla, Gavin D. M. Jeffries & Aldo Jesorka
PIPETTES, METHODS OF USE, AND METHODS OF STIMULATING AN OBJECT OF INTEREST
Alar Ainla, Owe Orwar & Aldo Jesorka PCT/IB2010/003307. Priority: Dec 3rd
2009.
MICROFLUIDIC DEVICE WITH HOLDING INTERFACE AND METHODS OF USE
Alar Ainla, Gavin D. M. Jeffries, Owe Orwar & Aldo Jesorka PCT/US12/36758. Priority: May 6th
, 2011.
METHOD OF HYDRODYNAMIC MANIPULATION OF OBJECTS ATTACHED TO A TWO-DIMENSIONAL FLUID
Alar Ainla, Bodil Hakonen, Irep Gözen, Owe Orwar & Aldo Jesorka US Provisional Patent Application. Priority: July 30th
, 2012.
METHOD TO FABRICATE, MODIFY, REMOVE AND UTILIZE FLUID MEMBRANES
Alar Ainla, Irep Gözen & Aldo Jesorka
US Provisional Patent Application. Priority: January 19th , 2013.
Abbreviations
AFM Atomic force microscopy
AMPA 2-amino-3-(3-hydroxy-5-methyl-isoxazol-4-yl) propanoic acid AOBS Acousto-optical beam splitter
APD Avalanche photodiode APTES 3-aminopropyl triethoxy silane CE Capillary electrophoresis
CLSM Confocal laser scanning microscopy COC Cyclic olefin copolymer
COP Cyclic olefin polymers DNA Deoxyribonucleic acid DPN Dip-pen nanolithography
DQN Diazoquinone
DRIE Deep reactive-ion etching
FCS Fluorescence correlation spectroscopy FDP Fluorescein diphosphate
FEM Finite element method
FRAP Fluorescence recovery after photobleaching FRET Förster resonance energy transfer
GABA γ-Aminobutyric acid
GFP Green fluorescent protein HCF Hydrodynamically confined flow HF Hydrofluoric acid
LTI Linear time invariant
MEMS Microelectromechanical system NA Numerical aperture
PC Polycarbonate
PDE Partial differential equation PDMS poly(dimethylsiloxane) PMMA poly(methyl methacrylate) PMT Photomultiplier tube PWFM Pulse-width flow modulation PWM Pulse-width modulation
SICM Scanning ion-conductance microscopy SOI Silicon on insulator
TEOS Tetraethyl orthosilicate
TIRF Total internal reflection fluorescence TPE Thermoplastic elastomer
Re Reynolds number Pe Peclet number St Strouhal number p Pressure Q Flow rate R Flow resistance G Flow conductance C Compliance c Concentration
Mathematical notations
Integral over closed surface
Integral over volume
⨂ Convolution
Contents
1. Introduction ... 3
2. Fundamentals of Microfluidics ... 7
2.1 Fluid Physics ... 7
2.1.1 Flow ... 7
2.1.2 Mass Transport & Diffusion ... 13
2.1.3 Temperature ... 20
2.2 Microfluidics versus Microelectronics... 21
2.2.1 Analogies ... 21
2.2.2 Circuits ... 23
2.2.3 Differences ... 26
3. Methods ... 29
3.1 Fabrication of Microfluidic Devices ... 29
3.1.1 Additive Techniques ... 31 3.1.2 Subtractive Techniques ... 31 3.1.3 Forming Techniques ... 33 3.1.4 Channel Sealing... 35 3.1.5 Other Methods ... 36 3.1.6 Materials ... 36 3.1.7 Photolithography ... 38 3.1.8 PDMS Microfluidics ... 41 3.2 Microscopy ... 44 3.3 Simulations ... 49 4. Technology ... 51 4.1 Microfluidic Dilution ... 51
4.2.2 Solution Exchange - Need for Speed ... 59
4.2.3 Other Methods to Deliver Chemicals ... 61
4.3 Functional Biomembranes ... 62
4.3.1 Structure of the Biomembrane ... 62
4.3.2 Properties of the Biomembrane ... 63
4.3.3 Supported Lipid Membrane Technologies ... 63
Summary ... 65
Acknowledgements ... 69
References ... 71
Appendix Papers I-VII
1. Introduction
Squeezing flasks and tubes from chemistry labs into small chips filled with networks of channels, valves, mixers and reaction chambers is a goal of the new and rapidly blooming field of Microfluidics. These so called ‘Lab on a Chip’ devices may eventually revolutionize medical diagnostics as well as chemical and biological analysis and research.
-- In the 23rd
century, interplanetary travel has become as common as flights from London to Paris a few centuries ago. Of course, mankind hasn’t made progress only in rocket science. A traveler has great need for protection, while wandering in the vastness of space. That's why Starfleet is equipped with Tricorders [1]– handheld devices which can help, while scouting on an alien planet or examining the health of a person, to detect infections by space bugs. -- With this vision of the future depicted in the 1960's cult series Star Trek, director Gene Roddenberry was mere decades ahead of his time. It was in the 1980s, when a microfabricated gas-chromatography column was actually putting forward the first steps towards miniaturization in chemical analysis [2]. And it took yet another decade, until in 1991, the Swedish company Pharmacia Biosensor AB (later Biacore AB, now part of GE Healthcare) coined the name ‘Microfluidics’ in one of their scientific papers [3]. Despite of this little known origin, the term ‘Microfluidics’ has now become synonymous with an entire field of science and technology, which is focusing on liquid manipulation and chemistry inside microscale devices. The field has exploded during the last decade, which is indicted by a doubling of the number of scientific publications and patent applications in less than every three years (Figure 1.1 A). Of course, such intensive research has resulted in a multitude of achievements, including deeper understanding of fluid physics at small scales, different means of fabrication and control, numerous applications and already more than a hundred companies making microfluidics-related products. Still, when looking at the typical technology lifecycle model (Figure 1.1 B), microfluidics is in its puberty, undergoing rapid development toward maturation [4], which includes improvements in manufacturing as well as identification of new applications. This shall pave the way for the wide-scale use of the technology, eventually bringing benefits for the society in general, such as faster, cheaper and more comprehensive diagnostics [5]. While the interplanetary spaceships from the Star Trek world are still a matter of science fiction, the medical Tricorder, helping to tackle "earth bugs", is actually almost within our reach. In 2012 Qualcomm was announcing a Tricorder X Prize of 10 million USD for the team to successfully build a portable device which is able to detect autonomously 15 distinct and common diseases, and does it better or at least as good as a trained physician [6]. This device would allow to keep better track of
personal health, prevent diseases, and reduce queues in front of doctors’ doors. All of these points are of critical importance in the future, since the aging population will inevitably increase the social burden of healthcare needs.
Microfluidics is not only holding promises in diagnostics, but can also extend the technical possibilities in our chemical, biological and medical research laboratories, for example by increasing efficiency and throughput, or by providing equipment in size scales specifically fitted to address single cells or their parts.
The backbone of this thesis is indeed the development of microfluidics based research tools and components, providing biophysicists and biologists with new means to control the chemical environment around single-cells. The PhD project started with a computer controlled, general purpose microfluidic dilution device, designed to generate chemical waves with desired parameters (Paper I). When it turned out to be cumbersome to apply this initial device in real biological experiments, it inspired the development of the next concept, which we termed a microfluidic pipette (Paper II). This device allowed localized delivery of solutions at the size scale of single cells in open volumes. The concept was reshaped for improved usability, turning it into a multifunctional tool for bioscience research (Paper III). We have explored diverse uses of the device for single cell electroporation (Paper IV), delivery of neurochemicals to brain slices (Paper V), and temperature measurement (Paper VI), and finally made a leap into new application areas and established a general method for the printing and manipulation of 2D nanofluidic circuits - a "lab on a membrane" (Paper VII).
The thesis provides background and context for the research described in the included papers. First relevant fluid physics and transport processes are discussed,
Figure 1.1. Development of microfluidics technology. (A) Growth of microfluidics during the last two decades. Based on scientific (ISI Web of Knowledge) and patent (espacenet) databases. (B) Microfluidics in the technology life cycle. Different phases of innovation, adaptation and economic impact [7-8].
followed by practical instructions and considerations for designing microfluidic systems, and then the methods relevant for the thesis, such as microfabrication, imaging with different microscopy tools, and modeling are listed. The final chapter provides an in-depth overview, and comparison of the specific technologies studied and developed in this thesis, which are dilution, delivery of chemicals to adherent cells, and lipid membrane manipulations.
The author hopes that the thesis will not only earn him an advanced degree, but also provides some inspiration for new students who are starting to explore the field of microfluidics, as well as some useful hints and guidance to consider before designing chips. Unfortunately, the relevance of the thesis could potentially be short lived, due to the rapid development of the field. On the other hand, that is exactly what is making it so exciting to work with!
2. Fundamentals
of
Microfluidics
Microfluidics and its big brother Microelectronics. Microelectronics has provided a plentitude of inspiration, and also tools for fabricating Microfluidics. Both have similarities even when it comes to physical laws. Yet why has Microfluidics not been able to repeate the glory of its older sibling? Why it has been so difficult to fully mimic electronic systems? These questions will be addressed in the following chapter, along with a brief exploration of the physical principles of microfluidics.
2.1 Fluid Physics
2.1.1 Flow
Transport of liquid in microscale channels is central for most microfluidic systems, including the ones described in this thesis. Therefore it is important to understand the basic driving forces, mechanisms and properties of microflows. The mechanical aspect of the flow, which is the motion of liquid, is described by classical mechanics and hydrodynamics. In order to derive such a flow equation, one can actually start from basic principles of mechanics known as Newton’s laws, named after Sir Isaac Newton, who formulated them in his work Philosophiae Naturalis Principia Mathematica at 1687. Most important are his 2nd
and 3rd
law (Eq 2.1 and Eq 2.2) relating acceleration and force acting on a body, and defining the mutuality of interactions.
= (2.1)
where is the force acting on a body, m is its mass and the acceleration caused by the force.
Which implies, that if one body affects another with force , then the second body affects the first back with force , which has the same magnitude, but acts in opposite direction.
While the original notion of Newton’s 2nd
law is well suited for describing falling apples and the motion of other solid objects, it is less convenient for handling liquid streams. For solid objects their coordinate and velocity is used to describe motion, while for fluids the velocity field becomes handier. It means that we are not looking at the velocity of one particular "fluid particle", but instead at the velocity of the fluid in a certain place in space. The important difference can be understood, if we imagine a constant flow rate in a tube, which means that the velocity field should be also constant. However, if the tube has a narrower region, the liquid needs to accelerate to pass it, since the flow velocity must be higher in the narrower region in order to maintain an overall constant flow rate. This means, that a particular fluid particle, which would always be a subject to Newton’s 2nd
equation, may experience acceleration, even if the velocity field is steady. This is illustrated in figure 2.1.
Lets express the acceleration of fluid particle ′ from the velocity field during an infinitely small time step . This acceleration would contain then two parts; one due to the change in the velocity field, and another due to the changing position of the particle in the steady field.
= ′= ( + ) − ( )+ ( ( + )) − ( ( )) (2.3)
Using the chain rule of differentiation on a spatial component, eq. 2.3 becomes
= ( )+ ( ) ( ) (2.4)
Since = ( , ), the partial derivative notation shall be used here, giving
Figure 2.1. Steady flow in a tube with shrinking diameter. The concept of a velocity field (blue) and the velocity of a particular fluid particle ′ (red). Even though the velocity field is steady, the particular fluid particle can accelerate during its journey.
= ( , )+ ( )= + ∙ ∇ (2.5) From the acceleration of the fluid particle and its mass , we calculate the force
= (2.6)
In contrast to solid objects, the density would be a more suitable descriptor of a fluid than the mass of some arbitrarily chosen particle:
= = + ∙ ∇ (2.7)
In most microfluidic systems where water solutions are used, it is safe to assume that the fluid is incompressible, which means that the density is a constant
( ) = 0 and = (2.8)
As seen from eq. 2.7, even a steady flow of a liquid features acceleration, and therefore forces acting between the fluid particles. The most common forces, which are always present, are caused by pressure and the viscosity of the fluids.
Liquid pressure acts on all surfaces and exerts a force in perpendicular direction to them. In order to derive the force exerted to a liquid particle with volume we need to integrate the pressure over the surface of this particle, where denotes a small surface element and points to its normal direction.
.= − ∙
(2.9)
By using Ostrogratsky’s divergence theorem, this integral over the surface can be turned into an integral over volume
.= − ∇
(2.10)
If the volume is infinitely small, we will have a differential form of the equation
= −∇ (2.11)
Where ∇ is the pressure gradient (also denoted grad(P) )
Another important force is friction inside fluid flow, which happens when different parts of the liquid move at different velocity. This was also studied by Newton, who found that the force required to overcome the friction and move two parallel plates
relative to each other, when both are separated by a fluid, is proportional to the velocity gradient (also called shear rate) / and the area of the plates (Figure 2.2A)
= (2.12)
with a proportionality constant , called dynamic viscosity, which is a characteristic property of the fluid, depending also on the temperature. However, this relation is correct only for some simple substances, called Newtonian fluids. In non-Netwonian fluids, like polymer solutions, depends also on shear rate, making the behavior much more complex. Fortunately water and most dilute aqueous solutions used in microfluidics behave as Newtonian fluids, therefore we focus only on them.
Starting from Newton’s law of viscosity we can derive how the viscous force inside the fluid is related to the velocity field. First lets imagine three close layers of fluid, which move with different speeds , and . If the layers are separated by distance , the forces acting between them can be calculated using Newton’s viscosity law (2.12)
= −
= −
(2.13)
Taking into account Newton’s 3rd
law about the mutuality of interactions (Eq. 2.2) and summing forces from both sides, we can calculate the net force acting on the middle layer of the fluid
= − = − − − (2.14)
As seen from this equation the force acting on the fluid layer does not depend on the velocity gradient, but on its change, which can in case of infinitely small be described through the derivative of the gradient.
Figure 2.2. Viscosity. (A) Two parallel plates separated by a Newtonian fluid, moving relative to each other (also known as Couette flow). (B) Relating viscous forces and velocity field.
= = (2.15) where is the volume of the layer. In order to generalized the equation, we calculate the force, considering infinitely small layers ( ) in three dimensions:
.= ∇ (2.16)
Now the different forces can be combined, and the net force is proportional to the acceleration
= .+ .
+ ∙ ∇ = −∇ + ∇
(2.17)
After dividing both sides by we obtain the following differential equation
+ ∙ = − + (2.18)
This equation, named after 19th
century French and British scientists Claude-Louis Navier and Sir George Gabriel Stokes (Navier-Stokes’ equation) relates the velocity field of the flow with the pressure distribution and viscosity. It is a central concept in all flow calculations. Unfortunately, this non-linear partial differential equation is almost never analytically solvable. In fact, understanding Navier-Stokes’ equation is considered one of the greatest unsolved problems in mathematics, with a million dollar price promised to anyone who can elucidate its properties [9].
But some qualitative insight into the flow behavior and its dependence on the size scales can be obtained by using scaling laws. For that purpose eq. 2.18 can be re-written in a dimensionless form, using the following replacements
= ̃ , ∇= 1 ∇ , = , = ̃ = (2.19)
Where ~ denotes dimensionless analogues. and are characteristic size and velocity scales.
̃+ ∙ ∇ = − ∇ + ∇
(2.20)
̃ + ∙ ∇ = −∇ + ∇
(2.21)
demonstrating that the dimensionless Navier-Stoke’ equation depends only on one scaling parameter, known as Reynolds number ( )
= (2.22)
If is large ( ≫ 1), the equation is dominated by the left side, which describes inertia. Due to the non-linear term ∙ ∇ the behavior of flow in a "high Reynolds number mode" is chaotic (turbulent flow). Alternatively, if is low ( ≪ 1), the inertial side can be neglected and the equation is dominated by pressure and viscosity terms. This linear equation, known as Stokes’ equation (Eq 2.23) has a well defined solution, which corresponds to the laminar flow regime. In typical microfluidic systems, where the channel sizes are small and the flow is slow, the Reynolds number is low and the flow is laminar.
∇ = ∇ (2.23)
Stokes' equation has analytical solutions for a variety of simple geometries like cylindrical and rectangular tubes. Due to fabrication constraints, microfluidic channels have most commonly rectangular geometry. In a long rectangular channel the flow field is [10] ( , ) =4ℎ ∆ 1 1 − ℎ ℎ ℎ 2ℎ , ℎ = = 0 (2.24)
Where is the channel length along the axis, ℎ is the channel height along the axis, is the channel width along the axis and ∆ is the pressure difference between
the channel ends. If the velocity is integrated over the cross section, the total flow rate can be found as = ( , ) ⁄ ⁄ =ℎ ∆ 12 1 − 1 192 ℎ ℎ 2ℎ , (2.25)
For practical purposes, this infinite series is still hard to calculate, therefore an approximation can be used
= ∆ ℎ
12 1 − 0.630 ℎ
where ℎ < . In worst case, when the channel has a square cross-section, the error generated by using this approximation is actually only 13%.
(2.26)
2.1.2 Mass Transport & Diffusion
From a chemical point of view, it would be boring to pump just water. Therefore, most of the microfluidic systems handle a variety of solutions and regents, are able to mix and switch between them, and carry out chemical reactions. This introduces a new dimension into the equations – the chemical composition. In case there is no reactivity between different fluids and no strong interaction between composition and flow, the composition can be split into independent concentrations of molecules and each of them can be considered separately. An example for a case where composition and flow behavior are not independent would be sugar solution and water, where the sugar concentration determines viscosity and therefore flow, which would then again influence concentration. This type of coupling makes calculations significantly more complex. In contrast, the research presented in this thesis involved only dilute solutions, where the chemical composition has no major effect on the flow properties. In this case the concentration of a substance can be described by the convection-diffusion equation (Eq. 2.27)
= − ∙ ∇ + ∇ (2.27)
where the left-hand side describes the temporal change of the concentration, which depends on the convective transport by the flow (− ∙ ∇ ), and on diffusion ( ∇ ). This equation has a striking similarity to the Navier-Stokes equation, where just has been replaced by . Therefore, the Navier-Stokes equation can be considered as the transport equation of momentum, where viscosity acts as diffusivity of momentum.
If we convert the convection-diffusion equation into a dimensionless form, as we did previously with Navier-Stokes (Eq. 2.21), the resulting equation depends on two scaling parameters and
̃
̃ = − ∙ ∇ ̃ + 1
∇ ̃ (2.28)
where stands for the Strouhal number, describing unsteadiness, and for Peclet number, describing the ratio between convective and diffusive mass transport. Pe is analogous to the Reynolds number, which describes the same for momentum.
= = (2.29)
where is the unsteady time.
In the following we consider two important cases, which are both also relevant for the research presented in this thesis. First, a steady flow and concentration patterns, where = 0, and second, transient propagation of concentration pulses in a pressure driven flow.
Steady flow
Our macro-world experience tells us that putting together two miscible liquids, for example syrup and water, will eventually result in their complete mixing. Microfluidics, on the other hand, offers an easy way to form and even maintain spatially constant concentration gradients. This requires that diffusion, which always mixes substances until the differences have faded, is compensated by the convective flow, which replaces mixed liquids. The dimensionless convection-diffusion equation (Eq. 2.28) shows that the stationary equation (St=0) depends on only one parameter, which is the Peclet number. Here, a higher Peclet number implies dominance of convection over diffusion, therefore less mixing and sharper concentration gradients, and vice versa. This is illustrated in figure 2.4, showing a typical T- or Y-channel, where two solutions enter a common channel, co-flow, and mix diffusively. The further we go from the junction point, the more diffusion has progressed, and the smoother is the gradient. A detailed description of the concentration profile is complex and requires numerical simulations. However, if we can assume a 2-dimensional channel with constant velocity, an analytical solution is possible (Eq. 2.30)
( , , ⁄ ) = 2 ⁄ − (2 + 1) ⁄ − ⁄ − 2 ⁄ ∈±ℕ (2.30)
where = ⁄4 is characteristic time-scale of the system, and = ⁄ is the time, during which diffusion has occurred. In reality this assumption is well suitable for high aspect ratio channels.
This principle has been used in a variety of devices, for example to generate concentration gradients for cell migration studies [11]. There is a class of separation techniques, based on devices called H-filters, where these two flows are split apart again at the end of the common channel [12]. The separation of the substances is based on their different diffusion properties. Even more efficient separation is achieved when active transport can be included and the selected substance can be dragged to one edge of the channel, for example by magnetic force [13], which has been used to remove pathogens from blood. The transport mode can be even biological. For example, similar filters have been used to separate live and dead sperms to improve in vitro fertilization [14]. Besides separation, this kind of dispersion in microchannels has to be considered when designing microfluidic mixers, to make sure that two fluids have been become well blended at the end of a mixing channel.
The same principle of convection competing with diffusion has found an application in hydrodynamic flow confinement (HCF), allowing localized delivery of chemicals. HCF is a key element of the multifunctional pipette, studied in this thesis.
Figure 2.4. Stationary concentration gradients in microchannels. (A) A Y-shaped channel, fed by two flows, where one is carrying a solution of a substance with concentration c0, and the other one is pure solvent (c=0). Due to the lack of turbulence, these two flows mix only due to molecular diffusion, which smoothens the concentration difference between the flows. On the other hand this diffusive mixing is compensated by a replenishing supply of liquids. The balance between diffusion and convective replacement is establishing a stationary concentration distribution (B). The concentration profile in this kind of channel depends only on one dimensionless time ratio t/T0. The greater the ratio, the smoother is the gradient.
Transient flow
But how will a fluid stream with unsteady composition be affected when it is transported in microfluidic channels? Or if we switch between different solutions? These questions can be answered by analyzing the transient propagation of concentration pulses. The convection-diffusion equation (Eq. 2.28) shows that in this case all three terms contribute to the equation, and an exact solution would depend on two dimensionless parameters, St and Pe. This makes it more complicated to formulate a universal description. However, depending on which phenomena are
Figure 2.5. Dispersion models and their regions of applicability depending on Pe number and channel geometries (L/r). (According to Probstein “Physiochemical Hydrodynamics” [15]). The colors used to shade the regions have blurry edges to emphasize that the transitions between these modes are not sharp. The illustrations show how a short plug of a substance is dispersed in different transport modes.
dominating it is possible to separate different transport modes, and provide simpler models to describe each of them [15]. This depends on two aspects, the Peclet number and the ratio of channel length and radius L/r (Figure 2.5). Here the Peclet number represents the ratio of convection along the channel axis (axial) and diffusion across its cross section (radial). But since the convection (∝ ) and diffusion (∝√ ) are scaling differently, the channel length has to be also considered in order to determine the right transport model.
Let’s look first at the case of a low Pe number, where the flow is slow and the diffusion is fast. If the channels are relatively short, the output is primarily dominated by diffusion (Pure axial diffusion). However, when the channel length is increased, the convection will eventually catch-up with diffusion, due to their different scaling. In this case, the convection dominates the axial transport, but the radial transport is still ruled by the diffusion, which means that the concentration over the cross-section of the channel is constant. (Axial convection, radial diffusion). the border between these modes is approximately at ≈ 0.4 ⁄ .
If we increase the flow and Peclet number, the radial diffusion cannot keep up with convection. In case of very fast flow and short channels, the pressure driven fluid stream is stretching the substance pulse into a parabolic shape, while diffusion does not have time for any significant action. Then the dispersion is only due to Pure convection. When the channels are made sufficiently long, both diffusion and convection are entering the process. Convection is stretching the concentration pulse and diffusion is mixing it in radial direction. This was studied by the British physicist Sir G. I. Taylor in the 1950's [16], who found that the interplay between axial stretching and radial diffusion causes the injected fluid plug to be smeared in the same way as diffusion does, but with a very much higher diffusion coefficient. This dispersion mode has been coined after him as Taylor dispersion. The effectively increased axial diffusion coefficient is called a Taylor dispersion coefficient
= 48
(2.31)
In contrast to the molecular diffusion coefficient , the Taylor dispersion coefficient is not a materials property, but depends on the geometries and flow rates in the tube. It is interesting to note that molecular diffusion has an inverse effect on the Taylor dispersion (Eq. 2.31). A higher D corresponds to lower dispersion. Pure Taylor dispersion neglects the axial diffusion, but when the Pe numbers are lower, both the contribution of Taylor dispersion and of molecular diffusion should be considered (Taylor-Aris dispersion), with the dispersion coefficient given by eq. 2.32.
= +
48
(2.32)
When it comes to practical calculations, it is important to notices that the above given coefficients (Eq. 2.31, 2.32) and boundaries between the dispersion modes are for
circular capillaries. In case of other channel shapes, geometry specific correction coefficients have to be used, while the scaling laws still hold universally. For example, for high aspect ratio channels with width , the Taylor-Aris dispersion becomes [17].
= +
210
(2.33)
In case of more complex geometries and transition region between the dispersion modes, it is often most efficient to use numerical computer simulations (discussed in a later chapter).
In all transport models (Figure 2.5) other than pure convection and the transition region neighboring it, the concentration can be considered constant across the channel cross-section (radially), and it varies only along the channel axis, i.e., dispersion and transport can be described as a one dimensional system.
This simplifies the mathematical representation. In all these cases, dispersion is described as diffusion, whether molecular or Taylor’s.
If we want to calculate how a solution of variable composition is affected by the transport through a channel, the notion of signals and system, where time dependent concentration takes the role of a signal, and diffusion in the channel takes the role of a system, can become useful. In this notion a system acts on the input signal, turning it into an output signal. The diffusion process is a linear-time invariant (LTI) system, meaning that i) changing concentration at the input would change the concentration at the output proportionally, and ii) the channel behaves exactly the same at different times. LTI systems have several useful properties. They can be described entirely by their impulse response function, which reflects how the system transforms an infinitely narrow input pulse (delta impulse). In case of diffusion, or Taylor dispersion, such a pulse would spread and evolve into a Gaussian.
( , ) = 1
√4 −4
(2.34)
where D is the dispersion coefficient and t is time. If we consider a channel with length and average velocity , the spatial coordinate can be turned into a time delay. = ( − )⁄ , giving eq. 2.35.
( ) = 1
√4 −
( − )
4
(2.35)
LTI implies that in this case the output signal of the system is a convolution of input signal and impulse response of the system.
( ) = ( )⨂ ( ) = ( ) ( − ) (2.36) LTI systems can be represented also in a frequency (Fourier) domain, where the convolution integral turns into a simple multiplication of the two spectra of the input signal and the impulse response.
( ) = ( ) ∙ ( ) (2.37)
Each spectrum can be found using the Fourier transform.
( ) = ( ) (2.38)
If we assume that axial convection is larger than dispersion, as it usually is, the impulse response of the channel would become
( ) ≈ 1
4 / −
( − )
4 /
(2.39)
which has Fourier transform
( ) ≈ 1
√2 − −
(2.40)
The magnitude of this spectrum is
( ) ∝ − = − (2.41)
This function has larger values in case of low frequencies ( ) and low values in case of higher, which means also that the channel acts as a chemical low-pass filter for concentration signals, letting to pass slow concentration waves, while damping sharp changes. Parameter can be referred as a cut-off frequency of the filter [18].
=
(2.42)
The cut-off frequency is higher with fast flow (no time for dispersion) and is lower with a higher dispersion coefficient and longer channels.
In the context of the research of this thesis, the chemical low-pass filter has been used in a microfluidic diluter (Paper I & II), where it smoothens fast pulses to a constant
concentration level. Note that if fast solution exchange is desirable (Multifunctional pipette (Paper III)), the dispersion effects are a limiting factor.
2.1.3 Temperature
Temperature is also affecting chemical and physical processes. It describes the motional energy of molecules, which affects their diffusion and reaction rate (Arrhenius law), as well as the chemical equilibria. Thermal transport is very similar to the convection-diffusion equation (Eq. 2.27)
= − ∙ ∇ + ∇ + (2.43)
where is the specific heat capacity, is the thermal conductivity, is the density and is the spatial heating power. Re-arranging the equation gives
= − ∙ ∇ + ∇ + (2.44)
where = / is the thermal diffusivity. Even though the thermal convection-diffusion equation is exactly the same as for the concentration case, there is one significant dissimilarity, which is especially important in the microfluidics realm: the difference of the diffusion constants between different materials. Molecules diffuse readily in liquids, but they do not enter into most solids (their diffusion constant is close to zero). Therefore we can consider that molecular diffusion occurs only inside channels, while , on the other hand, temperature is very similarly conducted by solids and liquids. For example, the thermal diffusivities in water and glass are 1.4 ∙ 10 ⁄ and 3.4 ∙ 10 ⁄ , respectively. This means that both the liquid and the device have to be considered when calculating heat transport in microfluidics. For practical purposes this is mostly done by using finite-element modeling (FEM). In comparison to molecular diffusion, thermal diffusion is much faster. This is favorable in case precise temperature control is needed, due to the fast thermal equilibration of the liquid to the device temperature. Fast diffusion makes it, on the other hand, harder to generate thermal gradients (Figure 2.4). In order to achieve a sufficiently high number, the channels have to be larger and flow faster. Nevertheless, thermal gradients established in microfluidic devices have been exploited to study, for example, developmental control mechanisms in fly embryos [19]. Thermal diffusion has been considered in the design of the optofluidic thermometer (Paper VI), where the flowrates had to be chosen to ensure confinement of fluorescent dyes, but would at the same time allow thermal equilibration.
2.2 Microfluidics
versusMicroelectronics
2.2.1 Analogies
Microfluidic and microelectronic device are not only fabricated in a similar way, they also hold similarities in the circuit theories used to describe them. With slight modifications, this analogy provides a variety of useful tools for designing and analyzing microfluidic circuits. Equation 2.26 describes the flow rate dependence in a microfluidic channel, where the flow rate is proportional to the pressure difference at the channel ends and to a parameter depending on channel geometry and viscosity. This is corresponding to Ohm’s law in electronics, which describes the proportionality between current and a voltage difference. Fluidic analogies for voltage, current, current density and charge would be pressure, flow rate, flow velocity and fluid volume, respectively. The channel geometry and viscosity dependent proportionality parameter is called "hydrodynamic resistance". Similar analogies exist also for other passive circuit elements capacitor and inductor (Figure 2.6). Hydrodynamic capacitance describes which volume of a liquid can be placed into a "liquid capacitor" per unit of pressure increase. In physical terms, the "liquid capacitor" can correspond to an elastic tube, which is enlarged in volume when pressurized, or to air-bubbles in a channel, which can be compressed. Inductance corresponds to mechanical inertia of the flow.
Table 2.1. Most common channel and tube geometries for microfluidics and their respective hydrodynamic resistances [10].
Geometry Channel resistance Figure
Circular =8 1 Rectangular ≈ 12 1 − 0.63(ℎ⁄ ) 1 ℎ Square = 28.4 1 Parabolic =105 4 1 ℎ
Figure 2.6. (On previous page) Analogies between electronics and fluidics.
Besides passive components, active components, like valves, are also commonly found in microfluidic systems, allowing modulation of the fluid flow. Pneumatic analogs for digital logics, latches and even fluidic processors have been reported [20-22].
2.2.2 Circuits
Similarly to electronics, these microfluidic elements can be combined to circuits with different properties. The most common of such circuits is a simple network of channels, which corresponds to a network of flow resistors, where pressures are applied to the inlets. In order to find the flow rates in such a system, Kirchoff’s rule, stating that the sum of flows to every circuit node has to be zero (incompressible fluid and channel), can be applied. For practical calculations we can first redraw the system in the way shown in figure 2.7, by grouping inlets/outlets and internal nodes of the circuits. The resistances of each flow resistor, and the pressures at the inlets, are known, thus the complete flow pattern can be calculated after the pressures at the internal nodes have also been found. This requires solving a linear equation system. To compose such an equation we consider linearity, which means that the flow in every resistor can be expressed as a superposition of two flows, each starting from a different end of the resistor. For mathematical simplicity we can replace the resistances by conductivities = 1⁄ , then = ∙ . Now we can write the equation for an internal node , using Kirchoff’s current rule: the sum of the outflows from the node to every other node, and the inflows from every other node to the node has to be zero. For the node this can be written as eq. 2.45.
− + + + = 0 (2.45)
This system of linear equations can be re-arranged into a matrix form,
− − ⋯ ⋮ ⋱ ⋮ ⋯ − − ⋮ = − ⋯ ⋮ ⋱ ⋮ ⋯ ⋮ (2.46)
= − (2.47) Other circuit elements, such as fluidic capacitors and inductors can be also incorporated in the calculation by replacing resistances by impedances with complex values.
In the following we consider a few important circuits for microfluidics design, where multiple different types of elements have been combined (Figure 2.8). Since these are not just resistor networks, their response has also a temporal component, which shall be considered while designing the fluidic switching systems. The first example is a tube or a channel, with elastic walls, or with a compressible fluid in it. Such a tube acts as a series of resistors and capacitors (Figure 2.8A), known also as a RC-line, which slows and delays any pressure signal applied through it. Pressure propagation in such a tube is described by the differential equation (Eq. 2.48)
= (2.48)
If we turn it into a dimensionless form, we can see that the equation depends only on one dimensionless parameter = . The higher the constant the longer and slower is the response. This is an important consideration in fluidic switching systems, such as the multifunctional pipette presented in paper III . Similar RC circuits are involved in pneumatic valves (Figure 2.8B), since deflecting the valve will require a certain volume of fluid (capacitance), which is transported through the channel (resistance).
Figure 2.7. Calculating flows in arbitrary channel networks. The network has inlets and outlets, with defined pressures , and internal nodes with pressures which shall be found. Each internal node can be connected with another internal node through a resistor with conductance , or with inlet through a resistor with conductance . Nodes that are not connected have zero conductances between them. Also = 0.
= 1 − (2.49) RC constant ( = ) determines how fast these valves can be actuated.
Here, faster valve closure can be achieved by using higher control pressure, since the necessary pressure level would be reached sooner. The RC constant itself, however, is independent of pressure. This has been a consideration in the design of the microfluidic diluter (Paper I). The third example involves pressure driven flow from a test tube (Figure 2.8C). In order to establish a flow, the test tube has to be pressurized, which involves a gas flow from the supply. In case of small pressures, ≈ ⁄ . The last example shows a microfluidic analogue to an RL circuit, which is describing the inertia of the flow (Figure 2.8D). The hydraulic inductance can be expressed as = 2 ⁄ , which in case of a circular tube would be = 4 ⁄3 . The time
Figure 2.8. Circuit elements in a microfluidic design. (A) Pressure propagation in an elastic tube. (B) Control of microvalves. (C) Controlling pressures in a supply tube. (D) Hydrodynamic inductance.
constant of such a circuit is = ⁄ , and it scales with the channel size as ∝ , meaning that the significance of inertia is dropping rapidly with shrinking dimensions.
2.2.3 Differences
Even though there are many analogies, fluidics is so far having difficulties to fully mimic highly integrated electronic circuits. This is due to a few, essential differences between these two domains of technology (Figure 2.9). The most important one is the way how information is carried and processed. In electronics, it is electric potential and current, carried by a single particle: the electron. In fluidics it is the pressure and the fluid flow, but with the exception of some micropneumatic control systems [22], interesting information is usually carried neither by the flow nor by the pressure, but by the chemical composition. In contrast to the electron, there are nearly infinite numbers of molecules and mixtures possible. This has an implication: in fluidics it is not enough to convey information by waves of potential, but liquid has to actually travel through the system, which is much more time consuming and makes it hard to efficiently connect different chips with macroscopic tubing. The ease of interconnectivity has been the foundation for applicability of microelectronics. Microchips, which contain many microscopic transistors and require complex and expensive manufacturing are universal building blocks that are easily connected by macroscopic wires and circuit boards to create a particular functionality. Therefore great applications emerged (literally) from garage projects, for example the personal
computer. Modular microfluidic constructors have been also developed, but comprised circuits with significantly larger channels (several 100s of μm) [23-24]. Another important difference is that the function of electronic circuits usually does not depend on the length of the interconnections (except very high frequency electronics), while in fluidics, the interconnections are also entangled with the properties of the circuit - their length, volume and fluidic resistance has to be considered. This restricts the flexibility of the assembly. It is typically needed that all functions of the system are incorporated into one chip, specifically created for a desired purpose. Therefore facile and low-cost methods to prototype and fabricate chips are needed to promote the development of new devices for various applications. Closest to this ideal is PDMS microfluidics, which has gained huge popularity among developers [25]. PDMS microfluidics is more thoroughly described in the third chapter. Hopefully, currently emerging rapid prototyping techniques (e.g. 3D printing) will lead to even simpler methods for designing and createing microfluidic systems.
3. Methods
Dust particles, common in our ambient atmosphere, can be significantly larger than microstructures, therefore microfabrication has to be performed in a highly clean environment - a cleanroom. Special overalls are required for operators to protect the samples from the biggest source of contamination – us. The photo shows the author in the MC2 cleanroom at Chalmers, holding a silicon master used to manufacture PDMS multifunctional pipettes.
3.1 Fabrication of Microfluidic
Devices
Most typical microfluidic devices are containing features that vary in size over 3-to-7 orders of magnitude: a cm scale chip, mm scale solution reservoirs and interface ports, and 100-10 μm scale channels. However, microfluidics can also bridge with nanofluidics, bringing the channel size down to the nanometer scale. It is hard to fabricate all of them with a single technique. Therefore, combinations of various tooling technologies are required [26]. For example, the main device in this work, the multifunctional pipette, has been fabricated using a mold which consists of two parts - a microfabricated master to define the microchannels, and a milled cavity to give the device its shape and define the solution reservoirs (Figure 3.1). The following section gives an brief overview of fabrication technologies for microfluidic devices.
Figure 3.1. Size scales of a microfluidic device, visualized on the example of the Multifunctional Pipette.
Key element of most current microfluidic devices is a network of small channels, where liquid handling occurs. Notable exceptions are droplets on surfaces, paper and thread microfluidics. Apart from rare techniques, which allow for generation of channels directly [27], most approaches to form close channels involve two common
steps - fabrication of a channel groove and subsequent sealing of its opened side with another layer of material (Figure 3.2). This procedure can be repeated to build multilayered channel structures. A large variety of techniques has been developed to fabricate channels and seal them in numerous different materials.
In general, fabrication techniques can be divided into three groups: additive, where materials are selectively added and gaps between them form channels; subtractive, where materials are selectively removed, and forming, where materials are shaped with the help of a template ("master").
3.1.1 Additive Techniques
The most common and well established way to define microstructures is through a photolithographic process (described in detail later), where a thin film of a light sensitive polymer (photoresist) is applied on a flat surface. A pattern is created by illuminating the photoresist through a photomask, followed by selective removal (development) of either exposed or unexposed parts of the resist. In most photolithographic applications, the forming resist pattern is used as a physical mask for further processing, such as deposition or etching. In microfluidics, the resist layers can be used directly as building material to define the device. Especially popular is the negative epoxy photoresist SU-8. Multiple SU-8 layers can be fabricated on top of each other, and also bonded thermally [28]. This method allows fabrication of high precision, chemically resistant devices. Shortcomings are the expensive materials and instrumentation required. While typical lithography is limited to thin solid coatings of photoresists, stereolithography, a related technique, (patented 1984 by C.W. Hull) [29], works in a liquid bath of photopolymer. This allows the sample to be moved vertically, while the strong optical absorption of the polymer ensures that the photoreaction occurs only in a very thin layer on the surface, which is scanned by a laser. By this layer by layer approach it is possible to build large 3D structures. This technology was expensive, and accessible only for industrial prototyping, until recently, when affordable 3D printers have become available [30]. Stereolithography has been also used to make high precision (about ~40μm) objects composed of multiple materials [31]. Similarly, multiphoton lithography is a technique to build 3D structures optically, but with significantly smaller feature size (< 1 μm) [32]. This very sharp point is achieved by non-linear absorption of a focused femptosecond laser. Other, inkjet-like 3D printing techniques, have found applications to build hydrogels for tissue scaffolds [33] and even assemble multiple cell types, which can be used in the future for regeneration of tissue [34]. Lately, customized low-cost 3D printed reactionware has been used for chemical synthesis [35]. It is a visionary example, showing the transition of 3D printing technology from model making to low-cost prototyping and manufacturing of functional scientific instrumentation. Similar developments would be highly desirable also for microfluidics. So far, low-cost methods lack the resolution, while high resolution multiphoton lithography is expensive, slow and restricted to small structures. However, affordable 3D printing is already available to make somewhat larger 'millifluidics', which can be applied to create interfaces to microfluidic devices [36]. It can be expected that 3D printing and related techniques will also revolutionize the fabrication of microfluidic devices, as it anticipated in other areas of manufacturing. In some people's opinion, the third industrial revolution will be based on 3D printing and mass customization [37].
3.1.2 Subtractive Techniques
The most commonly used subtractive technique in microfabrication is etching - a chemical removal of material. In order to define features, the substrate is typically patterned, using a photoresist. Development creates openings in the resist layer, which
define where the etching takes place. Depending on the chemical environment, etching procedures are divided into wet etching in solutions and dry etching using gases and plasmas [38]. The materials most commonly etched are inorganic, such as oxides, silicon and glass, the latter two being the most significant for practical microfluidic applications [39].
Glass is usually wet etched, using concentrated hydrofluoric acid (HF). With 48% HF, the isotropic etch rate at room temperature is about 8 μm/min [40]. Since glass is an attractive material for microfluidics, and there is a lack of alternative fabrication processes, HF etching is used widely. It has serious disadvantages, though, such as health risks, but also the isotropic etching profile, which can only produce shallow, wide channels with round edges. Since photoresists do not withstand HF, an evaporated sacrificial metal layer (e.g. Cr/Au) must be used as etching masks. This is an additional expensive process step.
Silicon has long been a standard material for the electronics industry. Its numerous processing techniques are well established for microelectronics fabrication. From the microfluidics and MEMS prospective, the most interesting technique is deep reactive-ion etching (DRIE). This process was patented by Robert Bosch GmbH 1992 [41]. It allows etching of very high aspect ratio structures with nearly vertical side walls. This dry plasma etching process uses two steps, nearly isotropic chemical etching with SF6,
followed by surface passivation with an inert plasma deposited fluoropolymer. This cycle is repeated, until the desired depth is reached [42]. In order to introduce directionality, the substrate is biased, which causes vertical bombardment of the passivation layer by ions. This removes the polymer layer selectively from the bottom of the structure, while having less effect on the side walls. In this way etching proceeds only from the bottom. Number and length of cycles are determining overall etch speed and smoothness of the side walls. DRIE can produce deep (100s of μm), high aspect ratio (>25), uniform (~5%) structures at moderate speed (~6 μm/min) and with good selectivity on silicon over the photoresist etch mask (>50:1) [42]. If higher uniformity is needed, DRIE can be performed on silicon on insulator (SOI) wafers (<1%). In this case the etching proceeds until it reaches a buried oxide layer, which defines the final etch depth. Disadvantage of SOI wafers is their very high cost (almost 10x as high as regular Si wafers [43]), but it can be an option to fabricate precision masters used for replica molding [44]. The DRIE process has been adopted also for glass, using CHF3 as etching gas, but it suffers from poor selectivity, and requires an
anodically bonded silicon wafer as an etching mask [45]. The aspect ratio is also much lower compared to silicon (wall angle of about 85°). Therefore, efforts to perform DRIE on glass are only justified in specific cases.
Laser ablation is a technique, where high power laser radiation is used to remove material [46-47]. The exact ablation mechanism depends on the laser and the substrate material, but in general it can be photochemical (UV lasers) or thermal (IR laser). Depending on the instrument, ablation can occur as a parallel process, where the entire structure is illuminated through a mask, or as a serial process, where the beam is scanned over the pattern to be written. Typical substrates are polymers, but
also glass has been laser ablated, even though it requires more care, to avoid thermal stress and cracking [48]. Laser processing can be combined with post processing, such as wet etching. Laser ablation has been used for both fabrication of channel networks as well as for additional process steps, such as creation of interconnects in multilayered PDMS devices [49]. Low-cost commercial cutters are based on CO2
lasers. They can produce ~100 μm channels with a surface roughness of a few μm in PMMA. Higher resolutions can be obtained with pulsed excimer, Nd:YAG or Ti:sapphire lasers, which can ablate in sub-μm steps [50]. Disadvantages of laser processing are channel roughness, and uncontrollable surface properties due to radiation damage and debris, which is deposited around the ablation site. Main advantage is the single-step fabrication of channel structures from design to finished chip, without the need for photolithography. This makes laser ablation a good candidate for rapid prototyping of microfluidic devices.
Mechanical machining is mostly associated with manufacturing of macroscopic goods, but micromilling can be equally competitive for microfluidic purposes [51-52]. For example, a high performance milling machine with 1 μm movement precision and 50-400 μm diameter carbide milling bits has been used to machine a brass molding master, featuring 20 μm wide and 400 μm high channel structures (aspect ratio 20), with vertical side walls having an average roughness of <100 nm. PMMA CE chips produced from this master had a similar separation performance compare to the ones fabricated with a LIGA-made master. If needed, the machined metal structures can be further smoothened, for example, by electrochemical polishing [53]. For engraving of channels, milling bits with a diameter down to 5 μm exists, even though for economical reasons their high price (>1'000kr/pcs) and wear rate should be considered [54]. Examples of milled microfluidic devices in aluminum with channel sizes ranging from 100 to 700 μm, have found use in, for example, chemical synthesis of polymers [55] and fluorescent microparticles [56]. Like with laser ablation, the advantage of milling is that the structures can be created from the design in a single process. Furthermore, one machine can produce features with greatly different heights, which is rarely possible with other techniques. Disadvantages are the low throughput, high cost of machinery, and rapid wear of tools.
3.1.3 Forming Techniques
The world of low cost disposable microfluidic devices is ruled by the different forming techniques for polymers. All of them share a common feature: a template, also called master or mold, is used to define the shape of a soft, or softened polymeric material, which is thereafter hardened, and released from the template. The molds themselves have to be manufactured using other techniques.
Casting (also reactive injection molding, or replica molding) is the simplest method, with minimal capital investment required to start. This is the reason why it has become popular for microfluidic chip fabrication in academic laboratories. Casting is used to create devices in thermosetting and elastomeric materials, starting with a liquid pre-polymer, which is poured onto the master or injected into the mould cavity,
followed by curing. Heat or UV-light cross-links the pre-polymer, turning it into hard plastics or rubber, which is then removed from the mold. A great advantage of casting is that it requires no special equipment other than the mold, which can be easily fabricated with high precision using lithographic patterning of photoresists (e.g. SU-8). Since the pre-polymer mixture is of low viscosity, it usually fills the mold without problems. The relatively long setting time makes the process robust and easy to reproduce manually. However, this last aspect turns into a major disadvantage when trying to use the technique for high throughput production, since the curing time can be on the order of hours, compared to injection molding, which can complete a cycle in seconds. All devices in the present work have been made using casting or reactive injection molding of PDMS.
Injection molding is a well established industrial method to manufacture macroscopic parts in thermoplastic materials. Similar to casting, a soft polymer mass is injected into a mold where it solidifies, adopting the shape of the mold. Instead of chemical reaction, the liquid/solid state of a thermoplastic is controlled by heating and cooling it around its glass transition temperature ( ). Since the required temperature change is relatively small, injection molding can have very short production cycles, limited only by thermal diffusion. Due to the viscosity of molten polymers, high injection pressures (~1000 bars) are required [57]. Many other aspects have to be considered when adopting injection molding for a new design, such as the temperatures of plastic and mold, injection rate and pressure, cooling times and holding pressures [58-59]. The design has to also consider the release of the final solid part. Thermal and flow "memory" of high molecular weight polymers can cause uneven shrinkage, stress and bending. In order to cope with this multi-dimensional optimization problem, specific softwares exist to model the injection molding process. Compared to "classical" injection molding on the macro scale, microinjection molding has its own additional requirements, like the use of an evacuated mold, cycling of the mold temperature (the VarioTherm process), used to avoid cooling of the polymer before it has managed to fill the small features. On one hand, very high costs for capital equipment, the fabrication of the mold cavity, and the required process optimization have rendered injection molding essentially inaccessible to academic researchers. On the other hand, low maintenance and material costs, high throughput and automated manufacturing, as well as a wide variety of different material, has made injection molding the method of choice for larger-scale industrial production of microfluidic devices.
Embossing (also nanoimprint) is, like injection molding, used to shape thermoplastic materials, but the extent of geometrical transformation of the material is less [60]. The process starts with the insertion of a plastics piece between two mold plates, where it is heated until softening (around ). Thereafter the plates are pressed together, shaping the softened material, which is then cooled to solidify. Microstructure imprinting can be performed in vacuum to avoid defects by trapped air. The material flow is in this case much smaller compared to injection molding, therefore it causes less stress. In general, hot embossing is significantly simpler than injection molding, but also with lower throughput and a more restricted range of usable geometries. Since high stress
is involved due to pressing, the masters have to be stronger (typically DRIE etched silicon or electroplated metal), than the photoresist patterns used for PDMS casting. This additional difficulty makes embossing less common for prototyping. Nevertheless, hot-embossing is an excellent choice for relatively easy reproduction of fine (sub μm) and high aspect ratio structures in thermoplastics.
In conclusion, which choice to make from the above mentioned techniques, depends on geometrical and material requirements. It is also an economic decision determined by the scale of production. Casting is suited for small series (up to 100s of pieces), followed by embossing and injection molding for large scale (in the >10'000 pcs range) [50]. From the materials perspective, casting is limited to curable polymers, while embossing and injection molding can be used with a wider variety of thermoplastics. From the aspect of geometry, casting and injection molding have high flexibility to replicate 3D structures at different size scales in a single process step, being only limited by release-related restrictions. Embossing is suited only for 2D layouts.
3.1.4 Channel Sealing
All of above-mentioned methods can only produce channel grooves. In order to form closed channels, they have to be sealed. Some techniques are reviewed below. A Conformal seal is formed when two flat and smooth surfaces are pressed against each other [61]. This method is very simple, but the resulting devices have generally lower pressure resistance and mechanical integrity. Advantages are the reversibility, which allows the device to be reused, or the ability to chemically or biologically pattern a surface with material from the channels [62-63]. Also, since no pre-bonding treatment is needed, it is easier to integrate microfluidics with a surface which is already covered with a (potentially sensitive) pattern.
Gluing is another well known way to bind surfaces together. In microfluidics, gluing requires precautions to avoid that the channels are contaminated, or even filled, by the glue. An efficient way to transfer liquid glue is stamping from a thin spin-coated layer. For example, UV-curable adhesive and PDMS has been used for sealing [64].
Plasma bonding is a technique to achieve chemical bonding between surfaces after activation with plasma, an ionized gas, which due to its high energy state is extremely reactive. Plasma bonding is a "clean" technique, without the risk of clogging. However, for bond formation to occur between two solid surfaces, smoothness is required to allow molecular level contact. Typically, plasma bonding is used with soft materials. The most widely applied example is oxygen plasma bonding of PDMS to PDMS and PDMS to glass [25], which has also been used to fabricate all devices in the current work. However, plasma treatment is sometimes not sufficient, and the activated surfaces have to be treated further to form a molecular monolayer with suitable bonding chemistry. For example, 3-aminopropyl triethoxy silane (APTES) can be used to assist the bonding of PDMS and PMMA [65] or PDMS and various metals, PP, PE and even Teflon [66]. Tetraethyl orthosilicate (TEOS) has been used for plasma bonding of PMMA-PMMA at low-temperatures [67].
